Deep Dive: How Merchants Can Reduce the Risk of False Positives Through AI and ML

The risk of digital fraud is difficult to overstate for merchants of all sizes, especially as digital engagement continues to grow and open more doors to fraudulent activity. eCommerce fraud losses are expected to top $20 billion this year alone, up from $17.5 billion in 2020, and merchants are scrambling to confirm that their security systems are up to the task of stopping cybercriminals — or at least persuading them to seek other targets.

Sometimes the cure can be worse than the disease, however, due to the side effects of false positives. Research suggests that up to 15% of card-not-present (CNP) transactions are falsely flagged as fraudulent, causing annual revenue losses of $118 billion. These false positives can cost merchants up to 75 times more than the fraud itself, both in the actual value of canceled transactions and in the opportunity cost of losing further business from customers who decide to abandon a merchant for a competitor.

The following Deep Dive explores the causes and consequences of false positives for merchants and examines which technologies hold the most promise for more accurately identifying fraud and avoiding the costly mistake of falsely accusing legitimate customers.

The Causes and Effects of False Positives

Most false positives result from rules-based automated fraud detection systems, which operate on simple yes/no logic to determine whether a given transaction is likely to be the work of a fraudster. For example, purchases above a certain dollar amount or too many purchase attempts from the same IP address within a limited window of time might be flagged. These rules-based systems have been shown to decline up to 30% of all nonfraudulent purchases — a massive rate of error.

One study from IBM showed that well over 90% of the fraud notifications generated by rules-based detection systems do not result in the creation of a suspicious transaction report, meaning that large numbers of legitimate customers are being declined without any notable increase in fraud detection.

The results of such a high volume of false positives can be devastating for businesses. Forty-two percent of customers report abandoning their carts when their payment method is declined, and 40% would opt for placing an order with a different company rather than trying to continue with the current purchase. This latter group includes 19% of customers in the $800,000 to $900,000 annual income bracket and 32% of those in the $1 million-plus bracket, market segments that merchants would kick themselves over losing. Studies predict false positives will cost merchants $386 billion annually by 2023 if current trends continue.

Maintaining the status quo when it comes to false positives is clearly not an option for merchants, as many are turning to artificial intelligence (AI)- and machine learning (ML)-based fraud detection systems to get their cybercrime prevention under control.

How AI and ML Can Reduce False Positives

AI- and ML-based systems hold several advantages over the rules-based fraud detection programs responsible for most of the false positives largely because of the ability of these advanced systems to learn and grow smarter over time. Instead of relying on absolutes to identify suspicious transactions, these smart systems look at each interaction holistically, taking every factor into account and combining all of them to construct a more specific fraud risk measurement. This calculation also can be tweaked by each organization to decline only those transactions that exceed a certain risk threshold, offering individual businesses the opportunity to balance the competing needs of customer convenience and fraud detection for themselves.

Evidence paints an even more convincing picture of AI and ML’s ability to detect fraud while reducing false positives. Teradata reported in a case study of Danske Bank that deploying AI reduced false positives by 60% while simultaneously improving real fraud detection by 50%. The developer predicted that this reduction in false positives could potentially rise to 80% as the system continues to learn and refine its detection model.

There likely always will be competing priorities when it comes to fraud detection and customer convenience, but all merchants can agree that false positives do not further either cause. AI- and ML-based systems could be valuable tools in stopping fraudsters while allowing legitimate customers to shop unhindered.